The cost of microprocessors has dropped drastically due to the benefits of mass production. Contrary to the drop in the cost of microprocessor hardware, the cost of software to support these microprocessors has risen. This inconsistency in the cost of implementing microprocessor controlled apparatuses has hampered their proliferation.
One major reason for the high cost of microprocessor software stems from the fact that programming methods previously used with large computers are now being used in microprocessors. As an example, consider the system where portions of assemblers and compilers are loaded into read/write memory from an input/output device. Next, source language is read from one device while partly processed output is written on another. Subroutines are then read in from another device and addresses are relocated so that all code required to run a job fits together in memory with no wasted words between.
Although the latter approach to programming made sense in the early days of computers when memory was prohibitively expensive, memory costs have dropped drastically with read only memory currently being the cheapest kind and available in very small capacities. In addition, microprocessors are being utilized in many direct control environments which require no writing or rereading of data. Thus, in many cases, input/output devices are only required for program development.
It is almost universally true that mass production achieves low unit cost by turning out large sums of identical products. In the case of microprocessor programs which are often implemented in read only memories (ROMs), the obvious way of carrying out mass production is to manufacture large quantities of ROMs with identical contents. The drawback to this approach lies in the prevalent method of linking various pieces of software together into a complete software system, which requires many address changes in each segment. Thus, one factor that has prevented the mass distribution of modular computer programs implemented in ROM form is that often a program instruction must known the absolute location in the memory system of another instruction in the same subprogram, for example, for branching. Therefore, if a computer manufacturer wanted to offer a catalog of subprograms in ROM form, he would have to contend with the desire of different users to select various offerings and locate the selected offerings at different actual locations in memory. It would not be feasible for the manufacturer to assign each subprogram a unique location as the total available memory locations would rapidly be exhausted. What is needed then, is a way to modify the address portion of certain of the ROM'ed instructions before they are used as addresses by the microprocessor.
One solution to the problem presented lies in the use of base registers. Base registers provide a hardware facility that relocates addresses as needed during program execution, making it unnecessary to change them within the program.
In the prior art, base registers have been designed into the microprocessor's CPU chip or added on by the system builder. The apparatus described by John A. Carroll in "Solving Mass-Produced ROM Programming Problems With Base Registers," Computer Design, August, 1977, pp. 99, proposes a base register system for solving the problem presented. The Carroll apparatus includes a set of base registers, and adder/selector circuitry located functionally between the CPU's address bus lines and the memory system address bus lines. Several bits of the CPU's address bus are used to select a base register, which is added to the remaining bits to produce the memory system address.
Although the Carroll apparatus represents an advance over the prior art, it appears to have several drawbacks. First, since it is undesirable to require the CPU to massage the address read from program memory before issuing it on its address bus, the base register select bits must be precoded into the mass produced program ROM. Since a user application will most likely require the use of several base registers, the number of which will increase with the number of subprogram modules used in the system, each of the ROMs must provide a sufficient number of bits of base-register select information to accommodate the largest foreseeable programming system and this base register address is not alterable since it is ROM'ed. Hence, the situation is similar to the original problem experienced, with the branch register address space substituted for the memory system address space. Although this drawback could be overcome by swapping the base register contents before beginning operation with a subprogram which uses that branch register, such a routine is expensive in system program overhead. Alternatively, the branch register select portion of the address could be provided by special hardware in each system, but this would defeat the objective of a mass produced system.
The second drawback of the Carroll apparatus is that the hardware required to perform the address relocation is located at or within the CPU. That is, either the branch register array and select/adder circuitry must be located on a special chip or it must be on the CPU chip, which may be the most valuable silicon real estate in the system. In the latter case, the requirement that all of the branch registers that any user might need must be sold to all users would impose a significant drain on the logic available on the CPU chip. In either case, locating the branch register array and adder/selector centrally presents additional problems in the implementation of multiprocessor systems in which multiple CPUs, each performing different tasks at any one time, share common system address buses, data buses, and program memory.
A second solution to the problem presented lies in the implementation of a "jump relative" instruction as one member of the microprocessor's instruction set. Basically, a jump relative instruction is a branch instruction with the object of the branch specifying a positive or negative number of storage words. The microprocessor adds the object of the jump relative instruction to the storage address the jump relative instruction is located at to come up with the effective address to be branched to.
The main drawback of the jump relative instruction is that since it is ROM'ed, the object of the instruction cannot be altered. Hence, if a first subprogram located in a first ROM required the use of a second subprogram located in a second ROM, the memory system address space within which the second ROM was located would have to correspond to the effective address specified by the ROM'ed jump relative instruction. This would prevent the user from choosing from a catalog of ROM contained subprograms and locating the subprograms at any available location in the microprocessor's memory system he chose.
Further, the availability of a jump relative instruction would not compensate for the other benefits associated with a base register. A base register can be used to store the location of RAM workspace the subprogram requires for storing or manipulating data. A base register may also be used for storing personality constants which preselect some subset of the subprogram's capability or for providing the addresses of input/output devices.
Additionally, since the jump relative instruction would have to be implemented on the CPU chip, the logic required to implement the instruction might cause a drain on the limited space available on the CPU chip.
It is a general object of the present invention to eliminate these and other drawbacks of the prior art by providing an improved device for relocating addresses as needed during program execution.
It is another object of the present invention to provide a device for relocating program addresses which is physically a part of the ROM package within which the program is located.
It is a further object of the present invention to provide a device for attachment to a microprocessor which allows the microprocessor user to utilize selected, mass produced, ROM contained subprograms without the need of customized software or hardware.
It is still a further object of the present invention to provide a device, the use of which will lower software costs associated with a mass produced microprocessor.
These and other objects, features and advantages of the present invention will become more apparent from the detailed description of the preferred embodiment when read in conjunction with the drawings.